See

discussions, stats, and author profiles for this publication at: https://www.researchgate.net/publication/257347180

Frequency detection in vortex flowmeter for

low Reynolds number using piezoelectric
sensor and installation effects

Article in Sensors and Actuators A Physical · September 2012

DOI: 10.1016/j.sna.2012.07.004

CITATIONS READS

8 304

3 authors, including:

Amit Agrawal S. V. Prabhu

Indian Institute of Technology Bombay Indian Institute of Technology Bombay
149 PUBLICATIONS 1,588 CITATIONS 132 PUBLICATIONS 1,143 CITATIONS

SEE PROFILE SEE PROFILE

All content following this page was uploaded by Amit Agrawal on 30 December 2016.

The user has requested enhancement of the downloaded file. All in-text references underlined in blue are added to the original document
and are linked to publications on ResearchGate, letting you access and read them immediately.
 Sensors and Actuators A 184 (2012) 78–85

Contents lists available at SciVerse ScienceDirect

Sensors and Actuators A: Physical

journal homepage: www.elsevier.com/locate/sna

Frequency detection in vortex flowmeter for low Reynolds number

using piezoelectric sensor and installation effects
A. Venugopal, Amit Agrawal, S.V. Prabhu ∗
Department of Mechanical Engineering, Indian Institute of Technology Bombay, Powai, Mumbai 400076, India

a r t i c l e i n f o a b s t r a c t

Article history: Piezoelectric sensors are one of the most widely used sensors for vortex flowmeter application due to
Received 12 March 2012 their low cost. Various researchers have employed piezoelectric sensor for this application. However, the
Received in revised form 2 July 2012 location of the sensor and the performance of vortex flowmeter under disturbed conditions are seldom
Accepted 3 July 2012
reported. In the present study, experimental investigations are conducted with water as the working
Available online 14 July 2012
medium in a circular pipe of diameter 52.5 mm. The optimum position of the piezoelectric sensor behind
the trapezoidal bluff body is found out be 0.85 times the width of the bluff body. A new algorithm based
Keywords:
on empirical mode decomposition and autocorrelation decay rate is suggested to identify the vortex
Piezoelectric sensors
Vortex flowmeter
shedding frequency under low Reynolds numbers flow condition. The performance of the flowmeter is
Trapezoid also evaluated under different disturbed flow conditions to quantify the sensitivity of the flowmeter. The
EMD disturbances studied are single 90◦ bend, gate valve, globe valve, and two 90◦ out of plane bends. The
Autocorrelation overall uncertainty in the Strouhal number is within ±1.71%.
© 2012 Elsevier B.V. All rights reserved.

1. Introduction presence of external disturbances is not satisfactorily addressed,

Vortex flowmeters are one of the most widely used flowme-
ters in various flow measurement sectors. Compared to differential
pressure devices, they offer large turndown ratio (minimum flow
rate/maximum flow rate) 1:20 and better accuracy (±0.5–1% of
the reading) [1,2]. Vortex flowmeters are cheaper than Coriolis
and multi-path ultrasonic flowmeters. Piezoelectric sensors are one
of the most widely used sensors for vortex flowmeter application
because of their relatively low cost and ease in fabrication [3]. The
sensor uses piezoelectric elements to detect strain in some mechan-
ical arrangement having sufficient area exposed to the differential
pressure of the vortices. These sensors detect time-varying forces
(lift) to measure the vortex shedding frequency. Most of the com-
mercially available vortex flowmeters use a piezoelectric crystal
embedded inside a mechanical structure placed behind the vortex
shedder. Piping vibration, flow pulsations and electrical interfer-
ences are inevitably present in real life applications. Under these
extreme harsh conditions, the output signal from the piezoelectric
sensor of the vortex flowmeter is a complex signal consisting of
many undesirable frequency components along with the main vor-
tex shedding signal. Hence, signal processing of the complex raw
signal is one of the most important and difficult tasks in design-
ing vortex flowmeters with piezoelectric sensor. The sensitivity of
a piezoelectric sensor under low flow rate conditions and in the
∗ Corresponding author. Tel.: +91 22 2576 7515; fax: +91 22 2572 6875/3480.
E-mail addresses: svprabhu@me.iitb.ac.in, svprabhu@iitb.ac.in (S.V. Prabhu).
and forms the subject of our investigation.
Venugopal et al. [4] comprehensively summarized the key issues
pertinent to the design of vortex flowmeter. Igarashi [5] per-
formed experiments with piezoelectric sensors embedded inside
various vortex shedder bodies. However, details of the type of the
piezoelectric sensor employed and the location of the sensor with
reference to the bluff body were not provided. Hence, it is difficult
to reproduce and extend their findings. Zheng et al. [6] conducted
an experimental study with a piezoelectric probe placed behind a
trapezoidal bluff body. The optimum location of the sensor behind
the bluff body was reported to be half the wavelength of the vor-
tex street. However, these studies were restricted to open channel
flows which do not include the influence of blockage. The optimum
location of the piezoelectric sensor placed behind the bluff body is
seldom reported in the literature.
Venugopal et al. [7,8] explored duct wall differential pressure
method in detail covering optimum bluff body shapes, location
of the wall pressure sensor and installation effects. Various two
dimensional bluff bodies (cylinder, triangle and trapezoid) and
axisymmetric bluff bodies (rings and cones) were investigated. The
bluff body with trapezoidal shape was found to be most appropri-
ate shape in terms of signal amplitude and deviation in Strouhal
number.
In the early stage, vortex signal processing methods were
based on power spectrum analysis. Ghaoud and Clarke [9] analyt-
ically simulated the vortex flowmeter signal and compared with
experimental results. Pre-filtering along with zero-crossing algo-
rithm was explored to extend the lower operating range of the

0924-4247/$ – see front matter © 2012 Elsevier B.V. All rights reserved.
http://dx.doi.org/10.1016/j.sna.2012.07.004
 A. Venugopal et al. / Sensors and Actuators A 184 (2012) 78–85 79

the empirical mode decomposition method. The vortex energy ratio

Nomenclature greater than 80% was dictated as the normal operating condition.
However, the sturdiness of these methods under real applications
A cross sectional area of the pipe (m2 ) where the amplitude of vibration signals are comparable to the
D diameter of the pipe (m) vortex signal needs to be demonstrated [14].
d width of the bluff body (m) Like other flowmeters, vortex flowmeter is sensitive to
f vortex shedding frequency (Hz) upstream disturbances. The flow field investigation downstream
Um mean velocity (m/s) of the disturbance and its impact on the performance of the vor-
P0 power of the vortex shedding frequency component tex flowmeter are studied by Miau et al. [15] and Laneville et al.
(W) [16] using cobra probe and hotwire anemometer to study the axial
Pi power of the other component contained in the flow distribution and swirl intensity at various upstream distance
interval i (W) to the flowmeter. Yang et al. [17] studied the effect of a perforated
Pt total energy of the spectrum (J) plate upstream to the vortex flowmeter as a source of flow distur-
P(w) energy content in frequency component (J) bance. They compared the performance of a vortex flowmeter with
P differential pressure (Pa) a piezoelectric sensor downstream of the bluff body and a commer-
PSD power spectral density (W/Hz) cial vortex flowmeter (Yokogawa YF105). The study revealed that
Rxx normalized autocorrelation function the piezoelectric sensor placed downstream of the bluff body was
Ust total uncertainty in St more sensitive to turbulence intensities due to upstream distur-
U1 repeatability bance as compared to the commercial flowmeter. However, there
U2 repeatability is no information available on the sensitivity of a vortex flowme-
U3 uncertainty in the frequency resolution ter downstream to a gate valve/globe valve/out of plane bends,
U4 uncertainty in the flow calibration system which are the major flow profile distortion elements in industrial
x streamwise coordinate (m) applications. Therefore, the objectives of the present study are:

Non-dimensional numbers 
Reynolds number Um D

ReD
 fd
 behind a trapezoidal bluff body,
Staverage average Strouhal number St = Um
Stmax maximum Strouhal number
Stmin minimum Strouhal number
 
Stmax +Stmin
Stmean 2 under various disturbed flow conditions.
Stfd Strouhal number under fully developed conditions
SNR signal to noise ratio
Greek symbols
 fluid density (kg/m3 )
 dynamic viscosity (Pa s)
ı deviation (in percentage)
 deviation from fully developed conditions (in per-
centage)
flowmeter. However, the signal output from a piezoelectric sensor
consists of hydrodynamic noise, piping vibration and other inter-
ferences. Hence, power spectrum based signal processing methods
sometimes fail to differentiate the vortex shedding signal from
other frequencies. In the recent past, Hilbert transform, which is
a time-frequency domain analysis, is widely used for vortex shed-
ding frequency estimation. Empirical mode decomposition is the
key step in the application of Hilbert–Huang transform proposed
in detail with theoretical basis by Huang et al. [10]. Sun et al. [11]
proposed using Hilbert–Huang transform to estimate the vortex
shedding frequency. The complex raw signal was first decomposed
into intrinsic mode functions (IMF) with the help of empirical
mode decomposition (EMD). The percentage error in the meter fac-
tor obtained by this method was four times lower as compared
to the Fourier transform based method at low flow rates. Zheng
et al. [12] proposed an algorithm based on Hilbert transform and
empirical mode decomposition for weak vortex signal. The residue
after every intrinsic mode decomposition was subjected to prob-
ability density function estimate. A probability density of 5% was
selected as the termination criteria for the shifting process. The
last residual component corresponds to the vortex shedding fre-
quency. The minimum Reynolds number covered with this method
was ReD = 6500. Sun and Zhang [13] proposed an energy ratio based
method for diagnosing the vortex flowmeter performance based on
1. Optimization of the location of piezoelectric sensor placed
2. Explore signal processing methods for the detection of vortex
shedding frequency at different Reynolds numbers,
3. Evaluate the sensitivity of the performance of vortex flowmeter
2. Experimental setup
A closed loop water circuit is built to conduct the experi-
ments in a circular pipe of diameter 52.5 mm. Appropriate test
sections are designed and fabricated for conducting experimental
investigations with a piezoelectric sensor. A high accuracy water
flow dynamic calibration facility utilizing dynamic weighing is
employed in the present study. The system is capable of handling
flow rate in the range of 0–550 m3 /h. The accuracy of the present
facility depends on the diameter of the pipe and mass flow rate
range. The collecting tank is equipped with an electronic plat-
form scale load cell at the base. The collecting tank capacity is
4000 kg. The system can respond to a time change of 0.1 s with 1 kg
resolution. The uncertainty in the velocity estimation for the min-
imum Reynolds number covered in the present system is 0.34%.
The details of the experimental setup and calibration facility are
shown in Fig. 1. A mixed flow type of pump (capacity 78 m and
114 m3 /h) is used for circulating water through the system from
an underground sump. The flow rate is controlled with the help of
gate valves as shown in Fig. 1.
The load cell is interfaced to a personal computer through an
RS232 port. Two butterfly valves (150 mm) are provided down-
stream of the collecting tank in order to drain or collect water in
the tank. Water from the tank is recirculated back to the sump. The
experiments are conducted in a circular pipe of diameter 52.5 mm.
The test section is provided with an upstream pipe length of 35D
to achieve fully developed inlet velocity profile. Flexible bellows
are mounted at the inlet to the upstream pipe and outlet to the
downstream pipe to dampen the pipe vibrations. A clamp on type
ultrasonic flowmeter is installed 10D downstream of the test sec-
tion for coarse adjustment of the mass flow rate. The output from
the sensor is logged to a personal computer with the help of a
Picoscope 2203. The bluff body employed in the present study is
trapezoidal in shape with a 72.5◦ included angle and a blockage
80 A. Venugopal et al. / Sensors and Actuators A 184 (2012) 78–85

Fig. 1. Test facility for vortex flowmeter. (1) Pump; (2) gate valve; (3) vortex flowmeter; (4) piezoelectric sensor; (5) ultrasonic flowmeter; (6) measuring tank; (7) platform
scale with shear beam load cell.

ratio of 0.27 as shown in Fig. 1 [7,8]. The blockage ratio is defined The Strouhal number is computed based on the average of these
as the ratio of bluff body width to pipe diameter. 44 sets of frequencies. The dynamic calibration facility is operated
for 10 min for a given Reynolds number. The linearity of the vortex
flowmeter is defined as:
2.1. Data reduction
Stmax − Stmin
For a given Reynolds number, the sensor signals are sampled Linearity (%) = × 100 (2)
2 × Stmean
at a rate of 1 kHz for 10 min. Subsequently, the sampled data are
divided into 44 blocks (each comprising 12,000 points) for Fourier
The deviation of the Strouhal number is defined as:
spectral analysis. The Strouhal number is calculated as
St − Staverage
fd %Deviation(ı) = × 100 (3)
St = (1) Staverage
Um

where f is the average frequency of 44 sets of readings for a given where Staverage is the average value of the Strouhal number for all
Reynolds number. Reynolds number covered.
 A. Venugopal et al. / Sensors and Actuators A 184 (2012) 78–85 81

0.260 Table 1
x/d = 0.28 Linearity and deviation in St for various locations of piezoelectric sensor.
0.250 x/d = 0.42
Location (x/d) ReD Linearity (%) ı
x/d = 0.56
x/d = 0.71 0.42 26 × 103 –3 × 105 5.69 7
0.240 x/d = 0.85 0.56 26 × 103 –3.1 × 105 5.25 7.3
x/d = 1 27 × 103 –3.1 × 105
St

0.71 1.95 2.75

0.230 0.85 25 × 103 –3.2 × 105 1.14 1.55
1 25 × 103 –3.2 × 105 1.68 2.22

0.220

about to shed. The vortex formation length decreases with increase

0.210
0 50000 100000 150000 200000 250000 in Reynolds number. In order to obtain good linearity the optimum
ReD
location should take care of low Reynolds number, where the vortex
formation length is large. Beyond the optimum location the vortex
Fig. 2. Variation of St with ReD for various locations of piezoelectric sensor.
suffers rapid decay and hence the signal amplitudes deteriorate
significantly.
The deviation of the Strouhal number under disturbed condi- The frequency computations are also performed with autocor-
tions is defined as: relation function with suitable low pass filter implementation for
St − Stfd lower velocities. The results obtained from DFT and autocorrela-
%Deviation from fully developed condition() = × 100 (4)
Stfd tion function for the optimum location of x/d = 0.85 are shown in
Table 2. The results show that the overall deviations in Strouhal
where Stfd is the average Strouhal number obtained under fully number values are smaller in the case of autocorrelation function.
developed conditions. The uncertainties in Strouhal number estimation can be improved
at lower Reynolds number by using autocorrelation function for
2.2. Uncertainty analysis vortex frequency estimation [18]. However, the autocorrelation
function yields misleading results when multiple frequencies are
The overall uncertainty associated with the Strouhal number is: present in the signal especially at low Reynolds number range.
2
USt = U12 + U22 + U32 + U42 (5) This essentially requires effective low pass filtering for detection
of vortex shedding frequency.
Ust – total uncertainty in St;
U1 – repeatability; 4. Estimation of vortex shedding frequency at low Reynolds
U2 – reproducibility; numbers
U3 – uncertainty in the frequency resolution;
U4 – uncertainty in the flow calibration system. The performance of a vortex flowmeter at low Reynolds num-
ber is susceptible to poor signal to noise ratio due to inherent weak
The repeatability of the vortex flowmeter with piezoelectric sen- vortex strength. The hydrodynamic noise, pipe vibration and other
sor with 109 sets of readings for 95% confidence level is 1.178%. The interferences contribute to a reasonably large percentage of the
minimum frequency measurable with the piezoelectric sensor is total energy harvested from the piezoelectric sensor. The output
7 Hz, which gives an uncertainty of 1.19% in frequency resolution. from the sensor is directly proportional to the square of the veloc-
The uncertainty in the velocity estimate is 0.34%. The reproducibil- ity as shown in Fig. 3. At low flow rates, the strength of the vortex
ity in the present case is assumed to be negligible as compared to signal is of the order of the noise. Therefore, it becomes difficult to
other uncertainties. Hence, the overall uncertainty in the Strouhal extract the buried vortex signal from noise, using only the power
number is 1.71%. spectrum based signal processing methods. This essentially puts
limitation on the operating range of the vortex flowmeter. Hence,
3. Optimization of the piezoelectric sensor location the effectiveness of signal processing algorithms is put under real
test under low Reynolds number conditions. Peng et al. [19] sug-
The piezoelectric sensor is one of the most widely used sen- gested dual bluff body design to improve the lower operating range
sors for vortex flowmeter application. However, the location of the of the flowmeter.
sensor behind the vortex shedder is seldom reported in the liter- The power spectrum obtained for different velocities under fully
ature. In this context, experimental investigations are performed developed condition is shown in Fig. 4. At high Reynolds num-
for various downstream locations of the piezoelectric sensor with bers, the power of vortex signal is clearly distinct. However, at
a trapezoidal bluff body. The orientation of the piezoelectric sen- low Reynolds numbers, multiple peaks are observed in the power
sor behind the bluff body is shown in Fig. 1. The flat portion of spectrum with reasonable energy contents. In the present case
the sensor is placed parallel to the flow direction to ensure maxi-
mum out with minimal flow distortion. The streamwise locations Table 2
covered in the present study is x/d = 0.28–1. The criteria selected Summary of results obtained from DFT and autocorrelation method.
for the optimum location are minimum value of linearity, ı and ReD St ı
turndown ratio. The performance of differential pressure devices is
DFT Autocorrelation DFT Autocorrelation
mainly reported in terms of ı. Hence, in the present study ı values
are reported along with linearity for direct comparison with differ- 3.1 × 105 0.221 0.219 −0.73 −0.65
ential pressure devices. The frequencies are computed based on the 2.4 × 105 0.221 0.220 −0.49 −0.40
1.8 × 105 0.222 0.220 −0.28 −0.24
Discrete Fourier Transform (DFT) method, suitably implemented in 1.3 × 105 0.222 0.221 −0.28 −0.07
Matlab. The St with ReD for all locations is shown in Fig. 2. The results 1 × 105 0.222 0.221 −0.13 0.10
are summarized in Table 1. The best linearity of ±1.14% is obtained 5.8 × 104 0.223 0.222 0.36 0.33
for x/d = 0.85 with 1:11 turndown ratio. The most appropriate loca- 2.8 × 104 0.226 0.223 1.55 0.93
Linearity (%) ±1.14 ±0.79
tion for the sensor is the point where the vortex is fully grown and is
82 A. Venugopal et al. / Sensors and Actuators A 184 (2012) 78–85

Sensor Output

Spectrum Analysis
(ΔP) α Um 2

ΔP Frequency ComputaƟon

Operating range
Y Output
No. of Peaks = 1
Vortex frequency

N
Noise
Velocity (Um) EMD

Fig. 3. Sensor output variation with flow velocity.

AutocorrelaƟon IMF
Decay rate
vibration frequency of 22 Hz of the pump operating at 1440 rpm
is observed in the power spectrum. The peak corresponding to the
vortex shedding frequency needs to be identified with care under
such conditions, as the vortex shedding frequency may not corre-
spond to the dominant peak. Hence, there is a need to investigate f = Min Decay
further on the nature and distinct features of each peak before rate IMF
arriving at a final conclusion. The peaks which are far-off from
the operating range of the flowmeter can be filtered out with a
band-pass filter.
For a reliable estimation of vortex frequency at low Reynolds Output
number (<30,000), an algorithm is suggested in the present work. Vortex frequency
The proposed algorithm is demonstrated as follows: the output
from the piezoelectric sensor is fed to a dc-offset corrected DFT
Fig. 5. Flow chart for vortex shedding frequency estimation algorithm.
method to compute the power spectrum. In case of only one dis-
tinct peak, the output is the vortex shedding frequency. However,
in case of multiple peaks with no harmonic relation between them, minimum operating frequency range of the meter. In the proposed
the signal is decomposed into intrinsic mode functions (IMF) with algorithm, only the IMFs which are visible in the power spectrum
the help of the empirical mode decomposition (EMD) method. The are selected for further analysis. This can be achieved by selecting
EMD method decomposes the signal into IMFs until the residues some threshold limit (1:10) for the amplitude of frequencies in the
are close to minimum tolerance value. The alternate criterion for power spectrum.
stopping the decomposition is to restrict the frequencies below the Xu et al. [20] proposed autocorrelation function to estimate
the vortex shedding frequency under mechanical vibration noise.
The mechanical vibration noise was a synthesized noise (electrical
f = 95.01Hz vibration test system) with a sharp peak in the power spectrum.
80 ReD = 3.1 × 105
PSD (W/Hz )

The ratio of autocorrelation coefficients of vibration signal and vor-

tex signal at two different time steps (same for both the signals) is
40
computed; the one with the smaller value is selected as the vortex
signal. This method is tested in the present study under real sig-
0
0 100 200 300 nal, involving piping applications. It was observed that simply the
Frequency (Hz) ratio of autocorrelation coefficients at fixed time step yields mis-
f = 31.43 Hz leading results. In the present case the minimum ratio of 0.005 is
ReD = 1 × 105 obtained for the IMF corresponding to frequency of 3.67 Hz, where
PSD (W/Hz )

2
as the ratio for vortex shedding frequency (8.33 Hz) is 0.239. The
reason for such results is attributed to the broad spectrum of other
interferences, which make the autocorrelation correlation function
0 to decay faster. The method is modified and used in combination
0 100 200 300
Frequency (Hz) with EMD. The flow chart for the modified algorithm is shown in
0.0008 Fig. 5. In the present study, the decay rate of the autocorrelation
f = 9.31 Hz
ReD = 3.0 × 104 coefficient is chosen as the criterion for differentiating the vortex
PSD (W/Hz )

shedding frequency from other frequency components.

0.0004
To demonstrate the algorithm, output from the piezoelectric
f = 21.33Hz sensor at ReD = 2.8 × 104 (the lowest Reynolds number employed
f = 49.33Hz
0 in the measurements) placed at the optimum location (x/d = 0.85)
0 100 200 300 is decomposed into IMFs with the help of EMD as shown in Fig. 6.
Frequency (Hz)
The normalized autocorrelation function peaks are shown in Fig. 8.
Fig. 4. Power spectrum at different ReD with piezoelectric sensor placed at x/d = 0.85. The IMF with the largest decay rate is selected as the vortex
 A. Venugopal et al. / Sensors and Actuators A 184 (2012) 78–85 83

6
4 Raw Signal ReD= 2.8 × 104
Voltage (mV)

2
0
-2
-4
-6
0 0.2 0.4 0.6 0.8 1
Time (s)
2
IMF1 f = 84.33 Hz
1
Voltage (mV)

-1

-2
0 0.2 0.4 0.6 0.8 1
Time (s)
1.2
IMF2 f = 49.33 Hz
0.8
Voltage (mV)

0.4
0
-0.4
-0.8
-1.2
0 0.2 0.4 0.6 0.8 1
Time (s)
1.5
IMF3 f = 21.33 Hz
1
Voltage (mV)

0.5
0
-0.5
-1
0 0.2 0.4 0.6 0.8 1
Time (s)

0.8 IMF4 f = 8.33 Hz

Voltage (mV)

0.4
0
-0.4
-0.8
0 0.2 0.4 0.6 0.8 1
Time (s)

Fig. 6. Empirical mode decomposition at ReD = 2.8 × 104 .

shedding frequency (8.33 Hz in the present case). The asymptotic

nature of the remaining IMFs is evident in Fig. 7. The time period Fig. 8. Various upstream disturbances chosen in this study.
of each cycle in the IMFs may not be the same as the EMD methods
does not ensure this. The frequency values reported in the present is clearly demonstrated. The lower operating range of the vortex
study are the average frequencies. flowmeter can be extended with this technique using only single
In the presents study the demerits of Fourier transform based piezoelectric sensor.
technique in comparison to EMD combined with autocorrelation
5. Investigations under disturbed flow conditions
0.6
ReD = 2.8 × 104 The sensitivity of the vortex flowmeter is also evaluated in
0.5
f = 84.33 Hz the presence of various upstream flow disturbances. The flow
f = 49.32 Hz disturbances covered in the present study include single 90◦ bend,
f = 21.33 Hz gate valve, globe valve and two 90◦ out of plane bends as shown in
0.4
f = 8.33 Hz Fig. 8. Influence of the distance between the upstream disturbance
f = 3.67 Hz and the vortex flowmeter is studied. The effect of the upstream
Rxx

0.3
distance is considered to be negligible as long as the performance
parameter  is within the uncertainty intervals. The results of a
0.2
single 90◦ bend and 90◦ out of plane bends at various upstream dis-
turbances are shown in Fig. 9 with an uncertainty band of ±1.71%.
0.1 The performance parameter ␨ values are positive for 5D upstream
length and with the increase in upstream length (10D and 15D)
0 they tend to move toward negative values. This indicated that close
0 2 4 6 8 10
Peak Index to the disturbance the estimated vortex frequencies are higher than
under fully developed conditions. However, for all upstream length
Fig. 7. Autocorrelation decay of IMFs at ReD = 2.8 × 104 . the ␨ values variation with Reynolds number shows similar trend
84 A. Venugopal et al. / Sensors and Actuators A 184 (2012) 78–85

2
5D
Single 90° Bend 10D
1
15D
ζ

0 Out of plane 5D
90° Bend 10D

-1

-2
0 100000 200000 300000 400000
ReD

Fig. 9. Effect of single 90◦ bend and out of plane 90◦ bend at various upstream Fig. 12. Minimum upstream length requirements for various disturbances.
distances to the vortex flowmeter.

2 P(w) – energy content in frequency component (W);

25% Closed Pt – total energy of the spectrum;
1
5D 50% Closed we – upper bound frequency in the spectrum-sampling rate.
75% Closed
0
The energy content of vortex shedding frequency is taken as ±2%
ζ

25% Closed
-1 10 D 50% Closed around the maximum frequency. The signal to noise ratio (SNR)
75% Closed value obtained under single 90◦ bend is positive and the spec-
-2 trum is distinct at 12 Hz. However, in the rest of the cases, multiple
peaks are observed. Hence, the SNR values are negative. This shows
-3
0 100000 200000 300000 400000 that three dimensional effects broaden the spectrum. The upstream
ReD length required for various disturbances are shorter compared to
most of the manufacturers recommended length. This is mainly
Fig. 10. Effect of gate valve for various gate openings at 5D and 10D to the vortex
because in the present study the upstream length requirement is
flowmeter.
quoted based on the uncertainty value of ±1.71%. Most of the man-
(i.e. decrease in  value with increase in Reynolds number). The ufacturers quote the linearity of the flowmeter as the accuracy and
results at 10D for single 90◦ bend and 90◦ out of plane bends are the uncertainties are seldom reported. Hence, a direct comparison
with uncertainty band of ±1.71%. The results for gate valve at 5D is not feasible.
and 10D upstream to the flowmeter are presented in Fig. 10. The
results show that the most severe case is encountered with 25% f = 12 Hz
0.008 ReD = 3.9 × 10 4
PSD (W/Hz)

opening of the gate valve. This requires at least 10D upstream

SNR = 2.07
length.
0.004
Globe valve is considered as one of the dominant flow distor-
tion element in industrial applications. The results for globe valve
0
under 100% and 50% of the maximum flow are shown in Fig. 11. 0 100 200 300
The minimum upstream length required for various disturbances Frequency (Hz)
are shown in Fig. 12. The signal quality and quality parameter under
f = 8 Hz
all disturbed conditions at 5D for lowest Reynolds number is shown 0.0004 ReD = 2.6 × 10 4
PSD (W/Hz)

in Fig. 13. The signal quality parameter used is the signal to noise SNR = - 4.02

ratio computation was suggested by Miau et al. [21] is: 0.0002

 1.02f
Po 0
SNR (dB) = 10 log10 , Po = P(w) dw 0 100 200 300
Pr 0.98f Frequency (Hz)
0.0016
 we
ReD = 2.8 × 104
PSD (W/Hz)

Pt = P(w) dw, Pr = Pt − Po
f = 9 Hz SNR = - 4.38
0 0.0008

0
2 5D 0 100 200 300
10D Frequency (Hz)
100% of Flow
15D 0.0016

ReD = 2.9 × 104

ζ

PSD (W/Hz)

0 5D
50% of Maximum SNR = - 2.91
10D 0.0008
Flow
15D f = 9 Hz

0
-2 0 100 200 300
0 100000 200000 300000 Frequency (Hz)
ReD
Fig. 13. Power spectrum for lowest Reynolds number at 5D downstream various
Fig. 11. Effect of globe valve for 100% and 50% of maximum flow at various upstream flow disturbance. (1) Single 90◦ bend; (2) gate valve 75% closed; (3) two 90◦ out of
distances to the vortex flowmeter. plane bends; (4) globe valve 50% of maximum flow.
 A. Venugopal et al. / Sensors and Actuators A 184 (2012) 78–85
6. Conclusions
Piezoelectric sensors are the cheapest available sensors for vor-
tex flowmeter applications. The location of the sensor behind the
bluff body is crucial in obtaining good linearity and turndown ratio.
However, the optimal sensor location is not available in the open
literature. At low Reynolds numbers, piping vibrations and electri-
cal interferences makes the detection of vortex shedding frequency
difficult. The signal processing algorithm should be robust enough
to identify the vortex shedding frequency and unambiguously dif-
ferentiate it from other undesirable frequencies. A new algorithm
to achieve this is proposed in the present work. The results from
the present study are summarized as:
1. The non-dimensional optimum location for the piezoelectric
sensor behind the trapezoidal bluff body is 0.85 with a linearity
of ±1.14% and a turndown ratio of 1:11.
2. The upstream length required for various upstream disturbances
with piezoelectric sensor with ±1.71% uncertainties are
a. single 90◦ bend – 10D
b. two 90◦ out of plane bends – 10D
c. gate valve 75% closed – 10D
d. globe valve 50% of maximum flow – 15D.
3. Empirical mode decomposition combined with autocorrelation
decay rate can be used for vortex frequency estimation. The
method is shown to be robust and works reliably even at low
Reynolds numbers (<30,000).
References
[1] New technology flowmeter studies, http://www.flowresearch.com/newtech
brochure.PDF.
[2] KROHNE OPTISWIRL 4070C Installation Manual 2010, http://www.krohne.
com/Vortex Flowmeters Start en.6.0.html.
[3] Z. Sun, H. Zhang, J. Zhou, Investigation of the pressure probe properties as the
sensor in the vortex flowmeter, Sensors and Actuators A 136 (2007) 646–655.
[4] A. Venugopal, A. Agrawal, S.V. Prabhu, Review on vortex flowmeter – designer
perspective, Sensors & Actuators A 170 (2011) 8–23.
[5] T. Igarshi, Flow resistance and strouhal number of a vortex shedder in a circular
pipe, Japan Society Mechanical Engineering 42 (1999) 586–595.
[6] D. Zheng, T. Zhang, Y. Hu, Experimental investigations of the location of a piezo-
electric probe in a vortex flow sensor, Measurement Science and Technology
18 (2007) 3777–3783.
[7] A. Venugopal, A. Agrawal, S.V. Prabhu, Influence of blockage and upstream dis-
turbances on the performance of a vortex flowmeter with a trapezoidal bluff
body, Measurement 43 (2010) 603–616.
[8] A. Venugopal, A. Agrawal, S.V. Prabhu, Influence of blockage and shape of a
bluff body on the performance of vortex flowmeter with wall pressure mea-
surement, Measurement 44 (2011) 954–964.
[9] T. Ghaoud, D.W. Clarke, Modeling and tracking a vortex flowmeter signal flow,
Measurement and Instrumentation 13 (2002) 103–117.
View publication stats
85
[10] N.E. Huang, Z. Shen, S.R. Long, The empirical mode decomposition and the
Hilbert spectrum for nonlinear and non-stationary time series analysis, Pro-
ceedings of the Royal Society of London (1998) 903–995.
[11] Z.Q. Sun, J.M. Zhou, P. Zhou, Application of Hilbert–Huang transformation to
denoising in vortex flowmeter, Journal of Central South University of Technol-
ogy 13 (2006).
[12] D. Zheng, T. Zhang, J. Xing, J. Mei, Improvement of the HHT method and appli-
cation in weak vortex signal detection, Measurement Science and Technology
18 (2007) 2709–2776.
[13] Z.Q. Sun, H.J. Zhang, Application of empirical mode decomposition based energy
ratio to vortex flowmeter state diagnosis, Journal of Central South University
of Technology 16 (2009) 154–159.
[14] J.J. Miau, C.C. Hu, J.H. Chou, Response of a vortex flowmeter to impulsive vibra-
tions, Flow Measurement and Instrumentation 11 (2000) 41–49.
[15] J.J. Miau, C.W. Wu, C.C. Hu, J.H. Chou, A study of signal quality of vortex
flowmeter downstream of two elbow out of plane, Flow Measurement and
Instrumentation 13 (2002) 75–85.
[16] A. Laneville, A. Strezelecki, P. Gajan, P. Hebrard, Signal quality of a vortex
flowmeter exposed to swirling flows, Flow Measurement and Instrumentation
4 (1993) 151–154.
[17] C.T. Yang, J.H. Shaw, M.C. Wu, Verification of a PZT vortex sensor in disturbed
pipe flows, in: Proceedings of the 2nd International Symposium on Instrumen-
tation Science and Technology, Jinan, China, August 2002.
[18] C.C. Hu, J.J. Miau, T.L. Chen, Determination of real-time vortex shedding fre-
quency by a DSP, Journal of the Chinese Society of Mechanical Engineers 27
(2006) 335–342.
[19] J. Peng, F. Xin, C. Ying, Flow measurement of by a new type vortex flow meter
of dual triangulate bluff body, Sensors and Actuators A: Physical 115 (2004)
53–59.
[20] K.J. Xu, Q.L. Luo, G. Wang, S.S. Liu, Y.B. Kang, Frequency-feature based
Antistrong-disturbance signal processing method and system for vortex
flowmeter with single sensor, Review of Scientific Instruments 81 (2010)
075104–075110.
[21] J.J. Miau, C.C. Yang, J.H. Hou, K.R. Lee, A T-shaped vortex shedder for vortex
flowmeter, Flow Measurement and Instrumentation 4 (1993) 259–267.
Biographies
Venugoapl A. was born in 1985. He received his B. Tech from NIT Raipur India (2006)
and M. Tech from IIT Bombay India (2008). He worked for Forbes Marshall Pvt. Ltd.
for two years on research projects related to flowmeters. At present he is pursuing
PhD studies from IIT Bombay. His research area includes bluff body wake dynamics,
flow metering and heat transfer.
Amit Agrawal obtained his B.Tech from IIT Kanpur in 1996. He worked at Tata
Motors, Pune for two years, before joining graduate studies at the University of
Delaware, USA. After obtaining his PhD in 2002, he spent about 2 years as postdoc-
toral fellow at the University of Newcastle, Australia. He is currently an associate
professor at IIT Bombay. His areas of interest are experimental, numerical and the-
oretical investigation of turbulent flows and flow at the microscales.
S.V. Prabhu is associate professor at the Department of Mechanical Engineering,
Indian Institute of Technology, Bombay, India. He graduated with a B.E. (Mech. Engg.)
first class with distinction from Mysore University in 1988, a Master of Technology
from National Institute of Technology, Surathkal in 1991 and a PhD from Indian
Institute of Technology, Bombay in 1998. His research interests are flowmetering,
heat transfer studies involving jet impingement, fire dynamics, renewable energy
(hydrokinetic turbines and wind turbines), gas turbine blade cooling and melting
and solidification of PCM and metals.